Inductive Inference

As our epistemic ambitions grow, the common and scientific endeavours are becoming increasingly dependent on Machine Learning (ML). The field rests on a single experimental paradigm, which consists of splitting the available data into a training and testing set and using the latter to measure how well the trained ML model generalises to unseen samples. If the model reaches acceptable accuracy, an a posteriori contract comes into effect between humans and the model, supposedly allowing its deployment to target environments. Yet the latter part of the contract depends on human inductive predictions or generalisations, which infer a uniformity between the trained ML model and the targets. The paper asks how we justify the contract between human and machine learning. It is argued that the justification becomes a pressing issue when we use ML to reach ‘elsewheres’ in space and time or deploy ML models in non-benign environments. The paper argues that the only viable version of the contract can be based on optimality (instead of on reliability which cannot be justified without circularity) and aligns this position with Schurz’s optimality justification. It is shown that when dealing with inaccessible/unstable ground-truths (‘elsewheres’ and non-benign targets), the optimality justification undergoes a slight change, which should reflect critically on our epistemic ambitions. Therefore, the study of ML robustness should involve not only heuristics that lead to acceptable accuracies on testing sets. The justification of human inductive predictions or generalisations about the uniformity between ML models and targets should be included as well. Without it, the assumptions about inductive risk minimisation in ML are not addressed in full.

This paper offers a probabilistic treatment of the conditions for argument cogency as endorsed in informal logic: acceptability, relevance, and sufficiency (RSA). Treating a natural language argument as a reason-claim-complex, our analysis identifies content features of defeasible argument on which the RSA conditions depend, namely: (i) change in the commitment to the reason, (ii) the reason's sensitivity and selectivity to the claim, (iii) one's prior commitment to the claim, and (iv) the contextually determined thresholds of acceptability for reasons and for claims. Results contrast with, and may indeed serve to correct, the informal understanding and applications of the RSA criteria concerning their conceptual (in)dependence, their function as update-thresholds, and their status as obligatory rather than permissive norms, but also show how these formal and informal normative approaches can in fact align.

The present paper reports on rapid perceptual learning of intonation contour categories in 9- to 11-year-old children and adults. Intonation contours are temporally-extended patterns that are expected to pose significant working memory challenges if represented in their full detail. We report that both children and adults form relatively abstract representations of intonation contours: previously encountered and novel exemplars are categorized into a category equally often, as long as distance from the prototype is controlled. However, we also document age-related differences in categorization performance. Given the same experience, adults form narrower categories than children, which may make it harder for them to recognize new instances of a contour as belonging to a known category. In particular, children are more likely to accept examples that lack some of the characteristic features of the prototype as belonging to the category. Children are also more likely to accept distortions of the prototype that are less similar to the prototype than the training examples. There are also age-related differences in feature weight: while adults pay more attention to the end of the contour, children pay approximately equal attention to the beginning and the end of the contour. The age range we examine appears to capture the tail-end of the developmental trajectory for learning intonation contour categories: there is a continuous effect of age on category breadth within the child group but the oldest children (older than 10;3) are largely adult-like. We discuss implications for theories of categorization, including their applicability to categories of linguistic units, and challenges posed by intonational patterns to various learner populations.